To be Closer: Learning to Link up Aspects with Opinions

Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 3899–3909 November 7–11, 2021. c⃝2021 Association for Computational Linguistics 3899 To be Closer: Learning to Link up Aspects with Opinions Yuxiang Zhou1∗, Lejian Liao1, Yang Gao1†, Zhanming Jie2,3 and Wei Lu2 1School of Computer Science and Technology, Beijing Institute of Technology 2StatNLP Research Group, Singapore University of Technology and Design 3ByteDance AI Lab yxzhou@bit.edu.cn, liaolj@bit.edu.cn, gyang@bit.edu.cn allan@bytedance.com, luwei@sutd.edu.sg Abstract Dependency parse trees are helpful for discov- ering the opinion words in aspect-based sen- timent analysis (ABSA) (Huang and Carley, 2019). However, the trees obtained from off- the-shelf dependency parsers are static, and could be sub-optimal in ABSA. This is be- cause the syntactic trees are not designed for capturing the interactions between opinion words and aspect words. In this work, we aim to shorten the distance between aspects and corresponding opinion words by learning an aspect-centric tree structure. The aspect and opinion words are expected to be closer along such tree structure compared to the standard dependency parse tree. The learning process allows the tree structure to adaptively corre- late the aspect and opinion words, enabling us to better identify the polarity in the ABSA task. We conduct experiments on five aspect- based sentiment datasets, and the proposed model significantly outperforms recent strong baselines. Furthermore, our thorough analy- sis demonstrates the average distance between aspect and opinion words are shortened by at least 19% on the standard SemEval Restau- rant14 (Pontiki et al., 2014) dataset1. 1 Introduction Aspect-based sentiment analysis (ABSA) (Pang et al., 2008; Liu, 2012) aims at determining the sentiment polarity expressed towards a particu- lar target in a sentence. For example, in the sentence “The battery life of this laptop is very long, but the price is too high”, the sentiment ex- pressed towards the aspect term “battery life” is positive, whereas the sentiment towards the as- pect term “price” is negative. Early research ef- forts (Wang et al., 2016; Chen et al., 2017; Liu and Zhang, 2017; Li and Lu, 2019; Xu et al., 2020a) *Work done when visiting SUTD. †Corresponding author. 1Our code is available at https://github.com/ zyxnlp/ACLT. Loving the harry potter movie marathon... Loving the harry potter movie marathon... root root (b) Learned aspect-centric Tree (Ours) (a) Dependency parse tree from spaCy. Dist(Loving, harry) = 4, Dist(Loving, potter) = 3 Dist(Loving, harry) = 1, Dist(Loving, potter) = 2 Figure 1: An example with different tree representa- tions in the Twitter dataset. “Dist” returns the number of hops between two words in the tree. Words marked in red and blue are aspect and opinion, respectively. focus on using an attention mechanism (Bahdanau et al., 2015) to model interactions between aspect and context words. However, such attention-based models may suffer from overly focusing on the fre- quent words that express sentiment polarity while ignoring low-frequency ones (Tang et al., 2019; Sun and Lu, 2020). Recent efforts show that the syntactic structures of sentences can facilitate the identification of sentiment features related to aspect …

Content truncated to 500 words